blob: 5022772906b5659c9abba7c8b20584b13d403c06 [file] [log] [blame]
Willy Tarreau59f98392012-07-06 14:13:49 +02001/*
2 * Connection management functions
3 *
4 * Copyright 2000-2012 Willy Tarreau <w@1wt.eu>
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 */
12
Willy Tarreaue1e4a612012-10-05 00:10:55 +020013#include <errno.h>
14
Willy Tarreau59f98392012-07-06 14:13:49 +020015#include <common/compat.h>
16#include <common/config.h>
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +010017#include <common/namespace.h>
Emmanuel Hocdet4399c752018-02-05 15:26:43 +010018#include <common/hash.h>
19#include <common/net_helper.h>
Willy Tarreau59f98392012-07-06 14:13:49 +020020
Willy Tarreauc5788912012-08-24 18:12:41 +020021#include <proto/connection.h>
Willy Tarreaudd2f85e2012-09-02 22:34:23 +020022#include <proto/fd.h>
Willy Tarreau5f1504f2012-10-04 23:55:57 +020023#include <proto/frontend.h>
Willy Tarreau2da156f2012-07-23 15:07:23 +020024#include <proto/proto_tcp.h>
Willy Tarreau2c6be842012-07-06 17:12:34 +020025#include <proto/stream_interface.h>
Emeric Brun4f603012017-01-05 15:11:44 +010026#include <proto/sample.h>
Willy Tarreau59f98392012-07-06 14:13:49 +020027
Emeric Brun46591952012-05-18 15:47:34 +020028#ifdef USE_OPENSSL
29#include <proto/ssl_sock.h>
30#endif
31
Willy Tarreaubafbe012017-11-24 17:34:44 +010032struct pool_head *pool_head_connection;
33struct pool_head *pool_head_connstream;
Willy Tarreau13e14102016-12-22 20:25:26 +010034struct xprt_ops *registered_xprt[XPRT_ENTRIES] = { NULL, };
Willy Tarreauf2943dc2012-10-26 20:10:28 +020035
Christopher Faulet32f61c02018-04-10 14:33:41 +020036/* List head of all known muxes for PROTO */
37struct mux_proto_list mux_proto_list = {
38 .list = LIST_HEAD_INIT(mux_proto_list.list)
Willy Tarreau2386be62017-09-21 19:40:52 +020039};
40
Willy Tarreauf2943dc2012-10-26 20:10:28 +020041/* perform minimal intializations, report 0 in case of error, 1 if OK. */
42int init_connection()
43{
Willy Tarreaubafbe012017-11-24 17:34:44 +010044 pool_head_connection = create_pool("connection", sizeof (struct connection), MEM_F_SHARED);
45 if (!pool_head_connection)
Olivier Houcharde2b40b92017-09-13 18:30:23 +020046 goto fail_conn;
47
Willy Tarreaubafbe012017-11-24 17:34:44 +010048 pool_head_connstream = create_pool("conn_stream", sizeof(struct conn_stream), MEM_F_SHARED);
49 if (!pool_head_connstream)
Olivier Houcharde2b40b92017-09-13 18:30:23 +020050 goto fail_cs;
51
52 return 1;
53 fail_cs:
Willy Tarreaubafbe012017-11-24 17:34:44 +010054 pool_destroy(pool_head_connection);
55 pool_head_connection = NULL;
Olivier Houcharde2b40b92017-09-13 18:30:23 +020056 fail_conn:
57 return 0;
Willy Tarreauf2943dc2012-10-26 20:10:28 +020058}
59
Willy Tarreau59f98392012-07-06 14:13:49 +020060/* I/O callback for fd-based connections. It calls the read/write handlers
Willy Tarreau7a798e52016-04-14 11:13:20 +020061 * provided by the connection's sock_ops, which must be valid.
Willy Tarreau59f98392012-07-06 14:13:49 +020062 */
Willy Tarreau7a798e52016-04-14 11:13:20 +020063void conn_fd_handler(int fd)
Willy Tarreau59f98392012-07-06 14:13:49 +020064{
Willy Tarreau80184712012-07-06 14:54:49 +020065 struct connection *conn = fdtab[fd].owner;
Willy Tarreau9e272bf2012-10-03 21:04:48 +020066 unsigned int flags;
Olivier Houchardaf4021e2018-08-09 13:06:55 +020067 int io_available = 0;
Willy Tarreau59f98392012-07-06 14:13:49 +020068
Willy Tarreaud80cb4e2018-01-20 19:30:13 +010069 if (unlikely(!conn)) {
70 activity[tid].conn_dead++;
Willy Tarreau7a798e52016-04-14 11:13:20 +020071 return;
Willy Tarreaud80cb4e2018-01-20 19:30:13 +010072 }
Willy Tarreau59f98392012-07-06 14:13:49 +020073
Willy Tarreau7d281492012-12-16 19:19:13 +010074 conn_refresh_polling_flags(conn);
Willy Tarreau916e12d2017-10-25 09:22:43 +020075 conn->flags |= CO_FL_WILL_UPDATE;
76
Willy Tarreau7d281492012-12-16 19:19:13 +010077 flags = conn->flags & ~CO_FL_ERROR; /* ensure to call the wake handler upon error */
Willy Tarreaud29a0662012-12-10 16:33:38 +010078
Willy Tarreauc76ae332012-07-12 15:32:13 +020079 process_handshake:
Willy Tarreauf9dabec2012-08-17 17:33:53 +020080 /* The handshake callbacks are called in sequence. If either of them is
81 * missing something, it must enable the required polling at the socket
82 * layer of the connection. Polling state is not guaranteed when entering
83 * these handlers, so any handshake handler which does not complete its
Willy Tarreaud6e999b2013-11-25 08:41:15 +010084 * work must explicitly disable events it's not interested in. Error
85 * handling is also performed here in order to reduce the number of tests
86 * around.
Willy Tarreauf9dabec2012-08-17 17:33:53 +020087 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +010088 while (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR))) {
Willy Tarreau310987a2014-01-22 19:46:33 +010089 if (unlikely(conn->flags & CO_FL_ERROR))
Willy Tarreau2c6be842012-07-06 17:12:34 +020090 goto leave;
Willy Tarreau59f98392012-07-06 14:13:49 +020091
Bertrand Jacquin93b227d2016-06-04 15:11:10 +010092 if (conn->flags & CO_FL_ACCEPT_CIP)
93 if (!conn_recv_netscaler_cip(conn, CO_FL_ACCEPT_CIP))
94 goto leave;
95
Willy Tarreau22cda212012-08-31 17:43:29 +020096 if (conn->flags & CO_FL_ACCEPT_PROXY)
97 if (!conn_recv_proxy(conn, CO_FL_ACCEPT_PROXY))
98 goto leave;
99
Willy Tarreau57cd3e42013-10-24 22:01:26 +0200100 if (conn->flags & CO_FL_SEND_PROXY)
101 if (!conn_si_send_proxy(conn, CO_FL_SEND_PROXY))
Willy Tarreau5f1504f2012-10-04 23:55:57 +0200102 goto leave;
Emeric Brun46591952012-05-18 15:47:34 +0200103#ifdef USE_OPENSSL
104 if (conn->flags & CO_FL_SSL_WAIT_HS)
105 if (!ssl_sock_handshake(conn, CO_FL_SSL_WAIT_HS))
106 goto leave;
107#endif
Willy Tarreauc76ae332012-07-12 15:32:13 +0200108 }
109
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200110 /* Once we're purely in the data phase, we disable handshake polling */
111 if (!(conn->flags & CO_FL_POLL_SOCK))
112 __conn_sock_stop_both(conn);
Willy Tarreauc76ae332012-07-12 15:32:13 +0200113
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200114 /* The connection owner might want to be notified about an end of
115 * handshake indicating the connection is ready, before we proceed with
116 * any data exchange. The callback may fail and cause the connection to
117 * be destroyed, thus we must not use it anymore and should immediately
118 * leave instead. The caller must immediately unregister itself once
119 * called.
Willy Tarreau2542b532012-08-31 16:01:23 +0200120 */
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200121 if (conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200122 return;
Willy Tarreau2542b532012-08-31 16:01:23 +0200123
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200124 if (conn->xprt && fd_send_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200125 ((conn->flags & (CO_FL_XPRT_WR_ENA|CO_FL_ERROR|CO_FL_HANDSHAKE)) == CO_FL_XPRT_WR_ENA)) {
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100126 /* force reporting of activity by clearing the previous flags :
127 * we'll have at least ERROR or CONNECTED at the end of an I/O,
128 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200129 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100130 flags = 0;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200131 if (conn->send_wait != NULL) {
132 conn->send_wait->wait_reason &= ~SUB_CAN_SEND;
133 tasklet_wakeup(conn->send_wait->task);
134 conn->send_wait = NULL;
135 } else
136 io_available = 1;
Olivier Houchard53216e72018-10-10 15:46:36 +0200137 __conn_xprt_stop_send(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200138 }
Willy Tarreau59f98392012-07-06 14:13:49 +0200139
Willy Tarreau57ec32f2017-04-11 19:59:33 +0200140 /* The data transfer starts here and stops on error and handshakes. Note
141 * that we must absolutely test conn->xprt at each step in case it suddenly
142 * changes due to a quick unexpected close().
143 */
144 if (conn->xprt && fd_recv_ready(fd) &&
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200145 ((conn->flags & (CO_FL_XPRT_RD_ENA|CO_FL_WAIT_ROOM|CO_FL_ERROR|CO_FL_HANDSHAKE)) == CO_FL_XPRT_RD_ENA)) {
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100146 /* force reporting of activity by clearing the previous flags :
147 * we'll have at least ERROR or CONNECTED at the end of an I/O,
148 * both of which will be detected below.
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200149 */
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100150 flags = 0;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200151 if (conn->recv_wait) {
152 conn->recv_wait->wait_reason &= ~SUB_CAN_RECV;
153 tasklet_wakeup(conn->recv_wait->task);
154 conn->recv_wait = NULL;
155 } else
156 io_available = 1;
Olivier Houchard53216e72018-10-10 15:46:36 +0200157 __conn_xprt_stop_recv(conn);
Willy Tarreau9e272bf2012-10-03 21:04:48 +0200158 }
Willy Tarreau2da156f2012-07-23 15:07:23 +0200159
Willy Tarreauc76ae332012-07-12 15:32:13 +0200160 /* It may happen during the data phase that a handshake is
161 * enabled again (eg: SSL)
162 */
Willy Tarreaud6e999b2013-11-25 08:41:15 +0100163 if (unlikely(conn->flags & (CO_FL_HANDSHAKE | CO_FL_ERROR)))
Willy Tarreauc76ae332012-07-12 15:32:13 +0200164 goto process_handshake;
165
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100166 if (unlikely(conn->flags & CO_FL_WAIT_L4_CONN)) {
Willy Tarreauf8deb0c2012-09-01 17:59:22 +0200167 /* still waiting for a connection to establish and nothing was
168 * attempted yet to probe the connection. Then let's retry the
169 * connect().
Willy Tarreau2da156f2012-07-23 15:07:23 +0200170 */
Willy Tarreau239d7182012-07-23 18:53:03 +0200171 if (!tcp_connect_probe(conn))
Willy Tarreauafad0e02012-08-09 14:45:22 +0200172 goto leave;
Willy Tarreau2da156f2012-07-23 15:07:23 +0200173 }
Willy Tarreau2c6be842012-07-06 17:12:34 +0200174 leave:
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100175 /* Verify if the connection just established. */
Willy Tarreau7bf3fa32017-03-14 20:19:29 +0100176 if (unlikely(!(conn->flags & (CO_FL_WAIT_L4_CONN | CO_FL_WAIT_L6_CONN | CO_FL_CONNECTED))))
177 conn->flags |= CO_FL_CONNECTED;
178
Willy Tarreau8e3c6ce2017-08-28 15:46:01 +0200179 /* The connection owner might want to be notified about failures to
180 * complete the handshake. The callback may fail and cause the
181 * connection to be destroyed, thus we must not use it anymore and
182 * should immediately leave instead. The caller must immediately
183 * unregister itself once called.
184 */
185 if (((conn->flags ^ flags) & CO_FL_NOTIFY_DONE) &&
186 conn->xprt_done_cb && conn->xprt_done_cb(conn) < 0)
187 return;
188
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100189 /* The wake callback is normally used to notify the data layer about
190 * data layer activity (successful send/recv), connection establishment,
191 * shutdown and fatal errors. We need to consider the following
192 * situations to wake up the data layer :
193 * - change among the CO_FL_NOTIFY_DATA flags :
194 * {DATA,SOCK}_{RD,WR}_SH, ERROR,
195 * - absence of any of {L4,L6}_CONN and CONNECTED, indicating the
196 * end of handshake and transition to CONNECTED
197 * - raise of CONNECTED with HANDSHAKE down
198 * - end of HANDSHAKE with CONNECTED set
199 * - regular data layer activity
200 *
201 * Note that the wake callback is allowed to release the connection and
202 * the fd (and return < 0 in this case).
Willy Tarreau2396c1c2012-10-03 21:12:16 +0200203 */
Olivier Houchardaf4021e2018-08-09 13:06:55 +0200204 if ((io_available || (((conn->flags ^ flags) & CO_FL_NOTIFY_DATA) ||
Willy Tarreau3c0cc492017-03-19 07:54:28 +0100205 ((flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) != CO_FL_CONNECTED &&
Olivier Houchard910b2bc2018-07-17 18:49:38 +0200206 (conn->flags & (CO_FL_CONNECTED|CO_FL_HANDSHAKE)) == CO_FL_CONNECTED))) &&
Olivier Houchard7505f942018-08-21 18:10:44 +0200207 conn->mux->wake && conn->mux->wake(conn) < 0)
Willy Tarreau7a798e52016-04-14 11:13:20 +0200208 return;
Willy Tarreaufd31e532012-07-23 18:24:25 +0200209
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200210 /* commit polling changes */
Willy Tarreau916e12d2017-10-25 09:22:43 +0200211 conn->flags &= ~CO_FL_WILL_UPDATE;
Willy Tarreauf9dabec2012-08-17 17:33:53 +0200212 conn_cond_update_polling(conn);
Willy Tarreau7a798e52016-04-14 11:13:20 +0200213 return;
Willy Tarreau59f98392012-07-06 14:13:49 +0200214}
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200215
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200216/* Update polling on connection <c>'s file descriptor depending on its current
217 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200218 * in CO_FL_WAIT_*, and the data layer expectations indicated by CO_FL_XPRT_*.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200219 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200220 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200221 */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200222void conn_update_xprt_polling(struct connection *c)
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200223{
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200224 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200225
Willy Tarreau3c728722014-01-23 13:50:42 +0100226 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200227 return;
228
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200229 /* update read status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200230 if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_XPRT_RD_ENA)) == CO_FL_XPRT_RD_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200231 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100232 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200233 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200234 else if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_XPRT_RD_ENA)) == CO_FL_CURR_RD_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200235 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100236 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200237 }
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200238
239 /* update write status if needed */
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200240 if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_XPRT_WR_ENA)) == CO_FL_XPRT_WR_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200241 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100242 f |= CO_FL_CURR_WR_ENA;
243 }
Olivier Houchard1a0545f2017-09-13 18:30:23 +0200244 else if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_XPRT_WR_ENA)) == CO_FL_CURR_WR_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200245 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100246 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200247 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100248 c->flags = f;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200249}
250
251/* Update polling on connection <c>'s file descriptor depending on its current
252 * state as reported in the connection's CO_FL_CURR_* flags, reports of EAGAIN
253 * in CO_FL_WAIT_*, and the sock layer expectations indicated by CO_FL_SOCK_*.
254 * The connection flags are updated with the new flags at the end of the
Willy Tarreau0ffde2c2012-10-04 22:21:15 +0200255 * operation. Polling is totally disabled if an error was reported.
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200256 */
257void conn_update_sock_polling(struct connection *c)
258{
259 unsigned int f = c->flags;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200260
Willy Tarreau3c728722014-01-23 13:50:42 +0100261 if (!conn_ctrl_ready(c))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200262 return;
263
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200264 /* update read status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100265 if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_SOCK_RD_ENA)) == CO_FL_SOCK_RD_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200266 fd_want_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100267 f |= CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200268 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100269 else if (unlikely((f & (CO_FL_CURR_RD_ENA|CO_FL_SOCK_RD_ENA)) == CO_FL_CURR_RD_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200270 fd_stop_recv(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100271 f &= ~CO_FL_CURR_RD_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200272 }
273
274 /* update write status if needed */
Willy Tarreau310987a2014-01-22 19:46:33 +0100275 if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_SOCK_WR_ENA)) == CO_FL_SOCK_WR_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200276 fd_want_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100277 f |= CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200278 }
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100279 else if (unlikely((f & (CO_FL_CURR_WR_ENA|CO_FL_SOCK_WR_ENA)) == CO_FL_CURR_WR_ENA)) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200280 fd_stop_send(c->handle.fd);
Willy Tarreauc8dd77f2012-11-05 17:52:26 +0100281 f &= ~CO_FL_CURR_WR_ENA;
Willy Tarreaue9dfa792012-09-01 17:26:16 +0200282 }
Willy Tarreau310987a2014-01-22 19:46:33 +0100283 c->flags = f;
Willy Tarreaub5e2cbd2012-08-17 11:55:04 +0200284}
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200285
Willy Tarreauff3e6482015-03-12 23:56:52 +0100286/* Send a message over an established connection. It makes use of send() and
287 * returns the same return code and errno. If the socket layer is not ready yet
288 * then -1 is returned and ENOTSOCK is set into errno. If the fd is not marked
289 * as ready, or if EAGAIN or ENOTCONN is returned, then we return 0. It returns
290 * EMSGSIZE if called with a zero length message. The purpose is to simplify
291 * some rare attempts to directly write on the socket from above the connection
292 * (typically send_proxy). In case of EAGAIN, the fd is marked as "cant_send".
293 * It automatically retries on EINTR. Other errors cause the connection to be
294 * marked as in error state. It takes similar arguments as send() except the
295 * first one which is the connection instead of the file descriptor. Note,
296 * MSG_DONTWAIT and MSG_NOSIGNAL are forced on the flags.
297 */
298int conn_sock_send(struct connection *conn, const void *buf, int len, int flags)
299{
300 int ret;
301
302 ret = -1;
303 errno = ENOTSOCK;
304
305 if (conn->flags & CO_FL_SOCK_WR_SH)
306 goto fail;
307
308 if (!conn_ctrl_ready(conn))
309 goto fail;
310
311 errno = EMSGSIZE;
312 if (!len)
313 goto fail;
314
Willy Tarreau585744b2017-08-24 14:31:19 +0200315 if (!fd_send_ready(conn->handle.fd))
Willy Tarreauff3e6482015-03-12 23:56:52 +0100316 goto wait;
317
318 do {
Willy Tarreau585744b2017-08-24 14:31:19 +0200319 ret = send(conn->handle.fd, buf, len, flags | MSG_DONTWAIT | MSG_NOSIGNAL);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100320 } while (ret < 0 && errno == EINTR);
321
322
323 if (ret > 0)
324 return ret;
325
326 if (ret == 0 || errno == EAGAIN || errno == ENOTCONN) {
327 wait:
Willy Tarreau585744b2017-08-24 14:31:19 +0200328 fd_cant_send(conn->handle.fd);
Willy Tarreauff3e6482015-03-12 23:56:52 +0100329 return 0;
330 }
331 fail:
332 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH | CO_FL_ERROR;
333 return ret;
334}
335
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200336int conn_unsubscribe(struct connection *conn, int event_type, void *param)
337{
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200338 struct wait_event *sw;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200339
340 if (event_type & SUB_CAN_RECV) {
341 sw = param;
342 if (sw->wait_reason & SUB_CAN_RECV) {
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200343 conn->recv_wait = NULL;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200344 sw->wait_reason &= ~SUB_CAN_RECV;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200345 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200346 __conn_xprt_stop_recv(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200347 }
348 if (event_type & SUB_CAN_SEND) {
349 sw = param;
350 if (sw->wait_reason & SUB_CAN_SEND) {
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200351 conn->send_wait = NULL;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200352 sw->wait_reason &= ~SUB_CAN_SEND;
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200353 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200354 __conn_xprt_stop_send(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200355 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200356 conn_update_xprt_polling(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200357 return 0;
358}
359
Olivier Houchard6ff20392018-07-17 18:46:31 +0200360int conn_subscribe(struct connection *conn, int event_type, void *param)
361{
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200362 struct wait_event *sw;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200363
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200364 if (event_type & SUB_CAN_RECV) {
Olivier Houchard4cf7fb12018-08-02 19:23:05 +0200365 sw = param;
366 if (!(sw->wait_reason & SUB_CAN_RECV)) {
367 sw->wait_reason |= SUB_CAN_RECV;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200368 conn->recv_wait = sw;
Olivier Houchard4cf7fb12018-08-02 19:23:05 +0200369 }
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200370 event_type &= ~SUB_CAN_RECV;
Olivier Houchard53216e72018-10-10 15:46:36 +0200371 __conn_xprt_want_recv(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200372 }
373 if (event_type & SUB_CAN_SEND) {
Olivier Houchard6ff20392018-07-17 18:46:31 +0200374 sw = param;
Olivier Houcharde1c6dbc2018-08-01 17:06:43 +0200375 if (!(sw->wait_reason & SUB_CAN_SEND)) {
376 sw->wait_reason |= SUB_CAN_SEND;
Olivier Houchardfa8aa862018-10-10 18:25:41 +0200377 conn->send_wait = sw;
Olivier Houcharde1c6dbc2018-08-01 17:06:43 +0200378 }
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200379 event_type &= ~SUB_CAN_SEND;
Olivier Houchard53216e72018-10-10 15:46:36 +0200380 __conn_xprt_want_send(conn);
Olivier Houchard6ff20392018-07-17 18:46:31 +0200381 }
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200382 if (event_type != 0)
383 return (-1);
Olivier Houchard53216e72018-10-10 15:46:36 +0200384 conn_update_xprt_polling(conn);
Olivier Houchard83a0cd82018-09-28 17:57:58 +0200385 return 0;
Olivier Houchard6ff20392018-07-17 18:46:31 +0200386}
387
Willy Tarreaud85c4852015-03-13 00:40:28 +0100388/* Drains possibly pending incoming data on the file descriptor attached to the
389 * connection and update the connection's flags accordingly. This is used to
390 * know whether we need to disable lingering on close. Returns non-zero if it
391 * is safe to close without disabling lingering, otherwise zero. The SOCK_RD_SH
392 * flag may also be updated if the incoming shutdown was reported by the drain()
393 * function.
394 */
395int conn_sock_drain(struct connection *conn)
396{
Willy Tarreaue215bba2018-08-24 14:31:53 +0200397 int turns = 2;
398 int len;
399
Willy Tarreaud85c4852015-03-13 00:40:28 +0100400 if (!conn_ctrl_ready(conn))
401 return 1;
402
403 if (conn->flags & (CO_FL_ERROR | CO_FL_SOCK_RD_SH))
404 return 1;
405
Willy Tarreaue215bba2018-08-24 14:31:53 +0200406 if (fdtab[conn->handle.fd].ev & (FD_POLL_ERR|FD_POLL_HUP))
407 goto shut;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100408
Willy Tarreaue215bba2018-08-24 14:31:53 +0200409 if (!fd_recv_ready(conn->handle.fd))
410 return 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100411
Willy Tarreaue215bba2018-08-24 14:31:53 +0200412 if (conn->ctrl->drain) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200413 if (conn->ctrl->drain(conn->handle.fd) <= 0)
Willy Tarreaud85c4852015-03-13 00:40:28 +0100414 return 0;
Willy Tarreaue215bba2018-08-24 14:31:53 +0200415 goto shut;
416 }
417
418 /* no drain function defined, use the generic one */
419
420 while (turns) {
421#ifdef MSG_TRUNC_CLEARS_INPUT
422 len = recv(conn->handle.fd, NULL, INT_MAX, MSG_DONTWAIT | MSG_NOSIGNAL | MSG_TRUNC);
423 if (len == -1 && errno == EFAULT)
424#endif
425 len = recv(conn->handle.fd, trash.area, trash.size,
426 MSG_DONTWAIT | MSG_NOSIGNAL);
427
428 if (len == 0)
429 goto shut;
430
431 if (len < 0) {
432 if (errno == EAGAIN) {
433 /* connection not closed yet */
434 fd_cant_recv(conn->handle.fd);
435 break;
436 }
437 if (errno == EINTR) /* oops, try again */
438 continue;
439 /* other errors indicate a dead connection, fine. */
440 goto shut;
441 }
442 /* OK we read some data, let's try again once */
443 turns--;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100444 }
445
Willy Tarreaue215bba2018-08-24 14:31:53 +0200446 /* some data are still present, give up */
447 return 0;
448
449 shut:
450 /* we're certain the connection was shut down */
451 fdtab[conn->handle.fd].linger_risk = 0;
Willy Tarreaud85c4852015-03-13 00:40:28 +0100452 conn->flags |= CO_FL_SOCK_RD_SH;
453 return 1;
454}
455
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100456/*
457 * Get data length from tlv
458 */
459static int get_tlv_length(const struct tlv *src)
460{
461 return (src->length_hi << 8) | src->length_lo;
462}
463
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200464/* This handshake handler waits a PROXY protocol header at the beginning of the
465 * raw data stream. The header looks like this :
466 *
467 * "PROXY" <SP> PROTO <SP> SRC3 <SP> DST3 <SP> SRC4 <SP> <DST4> "\r\n"
468 *
469 * There must be exactly one space between each field. Fields are :
470 * - PROTO : layer 4 protocol, which must be "TCP4" or "TCP6".
471 * - SRC3 : layer 3 (eg: IP) source address in standard text form
472 * - DST3 : layer 3 (eg: IP) destination address in standard text form
473 * - SRC4 : layer 4 (eg: TCP port) source address in standard text form
474 * - DST4 : layer 4 (eg: TCP port) destination address in standard text form
475 *
476 * This line MUST be at the beginning of the buffer and MUST NOT wrap.
477 *
478 * The header line is small and in all cases smaller than the smallest normal
479 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
480 * can safely use MSG_PEEK and avoid buffering.
481 *
482 * Once the data is fetched, the values are set in the connection's address
483 * fields, and data are removed from the socket's buffer. The function returns
484 * zero if it needs to wait for more data or if it fails, or 1 if it completed
485 * and removed itself.
486 */
487int conn_recv_proxy(struct connection *conn, int flag)
488{
489 char *line, *end;
Willy Tarreau77992672014-06-14 11:06:17 +0200490 struct proxy_hdr_v2 *hdr_v2;
491 const char v2sig[] = PP2_SIGNATURE;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100492 int tlv_length = 0;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200493 int tlv_offset = 0;
Willy Tarreaub406b872018-08-22 05:20:32 +0200494 int ret;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200495
496 /* we might have been called just after an asynchronous shutr */
497 if (conn->flags & CO_FL_SOCK_RD_SH)
498 goto fail;
499
Willy Tarreau3c728722014-01-23 13:50:42 +0100500 if (!conn_ctrl_ready(conn))
Willy Tarreauf79c8172013-10-21 16:30:56 +0200501 goto fail;
502
Willy Tarreau585744b2017-08-24 14:31:19 +0200503 if (!fd_recv_ready(conn->handle.fd))
Willy Tarreaufd803bb2014-01-20 15:13:07 +0100504 return 0;
505
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200506 do {
Willy Tarreaub406b872018-08-22 05:20:32 +0200507 ret = recv(conn->handle.fd, trash.area, trash.size, MSG_PEEK);
508 if (ret < 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200509 if (errno == EINTR)
510 continue;
511 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200512 fd_cant_recv(conn->handle.fd);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200513 return 0;
514 }
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100515 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200516 }
Willy Tarreaub406b872018-08-22 05:20:32 +0200517 trash.data = ret;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200518 } while (0);
519
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200520 if (!trash.data) {
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100521 /* client shutdown */
522 conn->err_code = CO_ER_PRX_EMPTY;
523 goto fail;
524 }
525
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200526 if (trash.data < 6)
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200527 goto missing;
528
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200529 line = trash.area;
530 end = trash.area + trash.data;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200531
532 /* Decode a possible proxy request, fail early if it does not match */
Willy Tarreau77992672014-06-14 11:06:17 +0200533 if (strncmp(line, "PROXY ", 6) != 0)
534 goto not_v1;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200535
536 line += 6;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200537 if (trash.data < 9) /* shortest possible line */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200538 goto missing;
539
David CARLIER42ff05e2016-03-24 09:22:36 +0000540 if (memcmp(line, "TCP4 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200541 u32 src3, dst3, sport, dport;
542
543 line += 5;
544
545 src3 = inetaddr_host_lim_ret(line, end, &line);
546 if (line == end)
547 goto missing;
548 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100549 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200550
551 dst3 = inetaddr_host_lim_ret(line, end, &line);
552 if (line == end)
553 goto missing;
554 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100555 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200556
557 sport = read_uint((const char **)&line, end);
558 if (line == end)
559 goto missing;
560 if (*line++ != ' ')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100561 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200562
563 dport = read_uint((const char **)&line, end);
564 if (line > end - 2)
565 goto missing;
566 if (*line++ != '\r')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100567 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200568 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100569 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200570
571 /* update the session's addresses and mark them set */
572 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
573 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = htonl(src3);
574 ((struct sockaddr_in *)&conn->addr.from)->sin_port = htons(sport);
575
576 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
577 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = htonl(dst3);
578 ((struct sockaddr_in *)&conn->addr.to)->sin_port = htons(dport);
579 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
580 }
David CARLIER42ff05e2016-03-24 09:22:36 +0000581 else if (memcmp(line, "TCP6 ", 5) == 0) {
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200582 u32 sport, dport;
583 char *src_s;
584 char *dst_s, *sport_s, *dport_s;
585 struct in6_addr src3, dst3;
586
587 line += 5;
588
589 src_s = line;
590 dst_s = sport_s = dport_s = NULL;
591 while (1) {
592 if (line > end - 2) {
593 goto missing;
594 }
595 else if (*line == '\r') {
596 *line = 0;
597 line++;
598 if (*line++ != '\n')
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100599 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200600 break;
601 }
602
603 if (*line == ' ') {
604 *line = 0;
605 if (!dst_s)
606 dst_s = line + 1;
607 else if (!sport_s)
608 sport_s = line + 1;
609 else if (!dport_s)
610 dport_s = line + 1;
611 }
612 line++;
613 }
Olivier Houchard53216e72018-10-10 15:46:36 +0200614 __conn_xprt_stop_recv(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200615
616 if (!dst_s || !sport_s || !dport_s)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100617 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200618
619 sport = read_uint((const char **)&sport_s,dport_s - 1);
620 if (*sport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100621 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200622
623 dport = read_uint((const char **)&dport_s,line - 2);
624 if (*dport_s != 0)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100625 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200626
627 if (inet_pton(AF_INET6, src_s, (void *)&src3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100628 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200629
630 if (inet_pton(AF_INET6, dst_s, (void *)&dst3) != 1)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100631 goto bad_header;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200632
633 /* update the session's addresses and mark them set */
634 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
635 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, &src3, sizeof(struct in6_addr));
636 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = htons(sport);
637
638 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
639 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, &dst3, sizeof(struct in6_addr));
640 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = htons(dport);
641 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
642 }
Willy Tarreau4c20d292014-06-14 11:41:36 +0200643 else if (memcmp(line, "UNKNOWN\r\n", 9) == 0) {
644 /* This can be a UNIX socket forwarded by an haproxy upstream */
645 line += 9;
646 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200647 else {
Willy Tarreau4c20d292014-06-14 11:41:36 +0200648 /* The protocol does not match something known (TCP4/TCP6/UNKNOWN) */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100649 conn->err_code = CO_ER_PRX_BAD_PROTO;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200650 goto fail;
651 }
652
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200653 trash.data = line - trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200654 goto eat_header;
655
656 not_v1:
657 /* try PPv2 */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200658 if (trash.data < PP2_HEADER_LEN)
Willy Tarreau77992672014-06-14 11:06:17 +0200659 goto missing;
660
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200661 hdr_v2 = (struct proxy_hdr_v2 *) trash.area;
Willy Tarreau77992672014-06-14 11:06:17 +0200662
663 if (memcmp(hdr_v2->sig, v2sig, PP2_SIGNATURE_LEN) != 0 ||
664 (hdr_v2->ver_cmd & PP2_VERSION_MASK) != PP2_VERSION) {
665 conn->err_code = CO_ER_PRX_NOT_HDR;
666 goto fail;
667 }
668
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200669 if (trash.data < PP2_HEADER_LEN + ntohs(hdr_v2->len))
Willy Tarreau77992672014-06-14 11:06:17 +0200670 goto missing;
671
672 switch (hdr_v2->ver_cmd & PP2_CMD_MASK) {
673 case 0x01: /* PROXY command */
674 switch (hdr_v2->fam) {
675 case 0x11: /* TCPv4 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100676 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET)
677 goto bad_header;
678
Willy Tarreau77992672014-06-14 11:06:17 +0200679 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
680 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_v2->addr.ip4.src_addr;
681 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_v2->addr.ip4.src_port;
682 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
683 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_v2->addr.ip4.dst_addr;
684 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_v2->addr.ip4.dst_port;
685 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200686 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100687 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET;
Willy Tarreau77992672014-06-14 11:06:17 +0200688 break;
689 case 0x21: /* TCPv6 */
KOVACS Krisztianefd3aa92014-11-19 10:53:20 +0100690 if (ntohs(hdr_v2->len) < PP2_ADDR_LEN_INET6)
691 goto bad_header;
692
Willy Tarreau77992672014-06-14 11:06:17 +0200693 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
694 memcpy(&((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr, hdr_v2->addr.ip6.src_addr, 16);
695 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_v2->addr.ip6.src_port;
696 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
697 memcpy(&((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr, hdr_v2->addr.ip6.dst_addr, 16);
698 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_v2->addr.ip6.dst_port;
699 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
KOVACS Krisztian7209c202015-07-03 14:09:10 +0200700 tlv_offset = PP2_HEADER_LEN + PP2_ADDR_LEN_INET6;
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100701 tlv_length = ntohs(hdr_v2->len) - PP2_ADDR_LEN_INET6;
Willy Tarreau77992672014-06-14 11:06:17 +0200702 break;
703 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100704
705 /* TLV parsing */
706 if (tlv_length > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200707 while (tlv_offset + TLV_HEADER_SIZE <= trash.data) {
708 const struct tlv *tlv_packet = (struct tlv *) &trash.area[tlv_offset];
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100709 const int tlv_len = get_tlv_length(tlv_packet);
710 tlv_offset += tlv_len + TLV_HEADER_SIZE;
711
712 switch (tlv_packet->type) {
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100713 case PP2_TYPE_CRC32C: {
714 void *tlv_crc32c_p = (void *)tlv_packet->value;
715 uint32_t n_crc32c = ntohl(read_u32(tlv_crc32c_p));
716 write_u32(tlv_crc32c_p, 0);
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200717 if (hash_crc32c(trash.area, PP2_HEADER_LEN + ntohs(hdr_v2->len)) != n_crc32c)
Emmanuel Hocdet115df3e2018-02-05 16:23:23 +0100718 goto bad_header;
719 break;
720 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +0100721#ifdef CONFIG_HAP_NS
722 case PP2_TYPE_NETNS: {
723 const struct netns_entry *ns;
724 ns = netns_store_lookup((char*)tlv_packet->value, tlv_len);
725 if (ns)
726 conn->proxy_netns = ns;
727 break;
728 }
729#endif
730 default:
731 break;
732 }
733 }
734 }
735
Willy Tarreau77992672014-06-14 11:06:17 +0200736 /* unsupported protocol, keep local connection address */
737 break;
738 case 0x00: /* LOCAL command */
739 /* keep local connection address for LOCAL */
740 break;
741 default:
742 goto bad_header; /* not a supported command */
743 }
744
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200745 trash.data = PP2_HEADER_LEN + ntohs(hdr_v2->len);
Willy Tarreau77992672014-06-14 11:06:17 +0200746 goto eat_header;
747
748 eat_header:
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200749 /* remove the PROXY line from the request. For this we re-read the
750 * exact line at once. If we don't get the exact same result, we
751 * fail.
752 */
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200753 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200754 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200755 if (len2 < 0 && errno == EINTR)
756 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200757 if (len2 != trash.data)
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100758 goto recv_abort;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200759 } while (0);
760
761 conn->flags &= ~flag;
Emeric Brun4f603012017-01-05 15:11:44 +0100762 conn->flags |= CO_FL_RCVD_PROXY;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200763 return 1;
764
765 missing:
766 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
767 * we have not read anything. Otherwise we need to fail because we won't
768 * be able to poll anymore.
769 */
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100770 conn->err_code = CO_ER_PRX_TRUNCATED;
771 goto fail;
772
773 bad_header:
774 /* This is not a valid proxy protocol header */
775 conn->err_code = CO_ER_PRX_BAD_HDR;
776 goto fail;
777
778 recv_abort:
779 conn->err_code = CO_ER_PRX_ABORT;
Willy Tarreau26f4a042013-12-04 23:44:10 +0100780 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
Willy Tarreau8e3bf692012-12-03 15:41:18 +0100781 goto fail;
782
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200783 fail:
Willy Tarreaud486ef52012-12-10 17:03:52 +0100784 __conn_sock_stop_both(conn);
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200785 conn->flags |= CO_FL_ERROR;
Willy Tarreaue1e4a612012-10-05 00:10:55 +0200786 return 0;
787}
788
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100789/* This handshake handler waits a NetScaler Client IP insertion header
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000790 * at the beginning of the raw data stream. The header format is
791 * described in doc/netscaler-client-ip-insertion-protocol.txt
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100792 *
793 * This line MUST be at the beginning of the buffer and MUST NOT be
794 * fragmented.
795 *
796 * The header line is small and in all cases smaller than the smallest normal
797 * TCP MSS. So it MUST always be delivered as one segment, which ensures we
798 * can safely use MSG_PEEK and avoid buffering.
799 *
800 * Once the data is fetched, the values are set in the connection's address
801 * fields, and data are removed from the socket's buffer. The function returns
802 * zero if it needs to wait for more data or if it fails, or 1 if it completed
803 * and removed itself.
804 */
805int conn_recv_netscaler_cip(struct connection *conn, int flag)
806{
807 char *line;
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000808 uint32_t hdr_len;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100809 uint8_t ip_v;
Willy Tarreaub406b872018-08-22 05:20:32 +0200810 int ret;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100811
812 /* we might have been called just after an asynchronous shutr */
813 if (conn->flags & CO_FL_SOCK_RD_SH)
814 goto fail;
815
816 if (!conn_ctrl_ready(conn))
817 goto fail;
818
Willy Tarreau585744b2017-08-24 14:31:19 +0200819 if (!fd_recv_ready(conn->handle.fd))
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100820 return 0;
821
822 do {
Willy Tarreaub406b872018-08-22 05:20:32 +0200823 ret = recv(conn->handle.fd, trash.area, trash.size, MSG_PEEK);
824 if (ret < 0) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100825 if (errno == EINTR)
826 continue;
827 if (errno == EAGAIN) {
Willy Tarreau585744b2017-08-24 14:31:19 +0200828 fd_cant_recv(conn->handle.fd);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100829 return 0;
830 }
831 goto recv_abort;
832 }
Willy Tarreaub406b872018-08-22 05:20:32 +0200833 trash.data = ret;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100834 } while (0);
835
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200836 if (!trash.data) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100837 /* client shutdown */
838 conn->err_code = CO_ER_CIP_EMPTY;
839 goto fail;
840 }
841
842 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000843 * CIP magic, header length or
844 * CIP magic, CIP length, CIP type, header length */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200845 if (trash.data < 12)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100846 goto missing;
847
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200848 line = trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100849
850 /* Decode a possible NetScaler Client IP request, fail early if
851 * it does not match */
Willy Tarreau55e0da62018-09-20 11:26:52 +0200852 if (ntohl(*(uint32_t *)line) != __objt_listener(conn->target)->bind_conf->ns_cip_magic)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100853 goto bad_magic;
854
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000855 /* Legacy CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200856 if ((trash.area[8] & 0xD0) == 0x40) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000857 hdr_len = ntohl(*(uint32_t *)(line+4));
858 line += 8;
859 }
860 /* Standard CIP protocol */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200861 else if (trash.area[8] == 0x00) {
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000862 hdr_len = ntohs(*(uint32_t *)(line+10));
863 line += 12;
864 }
865 /* Unknown CIP protocol */
866 else {
867 conn->err_code = CO_ER_CIP_BAD_PROTO;
868 goto fail;
869 }
870
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100871 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin72fa1ec2017-12-12 01:17:23 +0000872 * a minimal IP header */
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200873 if (trash.data < 20)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100874 goto missing;
875
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100876 /* Get IP version from the first four bits */
877 ip_v = (*line & 0xf0) >> 4;
878
879 if (ip_v == 4) {
880 struct ip *hdr_ip4;
David Carlier3015a2e2016-07-04 22:51:33 +0100881 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100882
883 hdr_ip4 = (struct ip *)line;
884
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200885 if (trash.data < 40 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100886 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000887 * IPv4 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100888 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000889 }
890 else if (hdr_ip4->ip_p != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100891 /* The protocol does not include a TCP header */
892 conn->err_code = CO_ER_CIP_BAD_PROTO;
893 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000894 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100895
David Carlier3015a2e2016-07-04 22:51:33 +0100896 hdr_tcp = (struct my_tcphdr *)(line + (hdr_ip4->ip_hl * 4));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100897
898 /* update the session's addresses and mark them set */
899 ((struct sockaddr_in *)&conn->addr.from)->sin_family = AF_INET;
900 ((struct sockaddr_in *)&conn->addr.from)->sin_addr.s_addr = hdr_ip4->ip_src.s_addr;
901 ((struct sockaddr_in *)&conn->addr.from)->sin_port = hdr_tcp->source;
902
903 ((struct sockaddr_in *)&conn->addr.to)->sin_family = AF_INET;
904 ((struct sockaddr_in *)&conn->addr.to)->sin_addr.s_addr = hdr_ip4->ip_dst.s_addr;
905 ((struct sockaddr_in *)&conn->addr.to)->sin_port = hdr_tcp->dest;
906
907 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
908 }
909 else if (ip_v == 6) {
910 struct ip6_hdr *hdr_ip6;
David Carlier3015a2e2016-07-04 22:51:33 +0100911 struct my_tcphdr *hdr_tcp;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100912
913 hdr_ip6 = (struct ip6_hdr *)line;
914
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200915 if (trash.data < 60 || trash.data < hdr_len) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100916 /* Fail if buffer length is not large enough to contain
Bertrand Jacquin67de5a22017-12-13 01:15:05 +0000917 * IPv6 header, TCP header */
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100918 goto missing;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000919 }
920 else if (hdr_ip6->ip6_nxt != IPPROTO_TCP) {
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100921 /* The protocol does not include a TCP header */
922 conn->err_code = CO_ER_CIP_BAD_PROTO;
923 goto fail;
Bertrand Jacquinb3875912017-12-13 00:58:51 +0000924 }
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100925
David Carlier3015a2e2016-07-04 22:51:33 +0100926 hdr_tcp = (struct my_tcphdr *)(line + sizeof(struct ip6_hdr));
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100927
928 /* update the session's addresses and mark them set */
929 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_family = AF_INET6;
930 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_addr = hdr_ip6->ip6_src;
931 ((struct sockaddr_in6 *)&conn->addr.from)->sin6_port = hdr_tcp->source;
932
933 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_family = AF_INET6;
934 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_addr = hdr_ip6->ip6_dst;
935 ((struct sockaddr_in6 *)&conn->addr.to)->sin6_port = hdr_tcp->dest;
936
937 conn->flags |= CO_FL_ADDR_FROM_SET | CO_FL_ADDR_TO_SET;
938 }
939 else {
940 /* The protocol does not match something known (IPv4/IPv6) */
941 conn->err_code = CO_ER_CIP_BAD_PROTO;
942 goto fail;
943 }
944
Bertrand Jacquin7d668f92017-12-13 01:23:39 +0000945 line += hdr_len;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200946 trash.data = line - trash.area;
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100947
948 /* remove the NetScaler Client IP header from the request. For this
949 * we re-read the exact line at once. If we don't get the exact same
950 * result, we fail.
951 */
952 do {
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200953 int len2 = recv(conn->handle.fd, trash.area, trash.data, 0);
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100954 if (len2 < 0 && errno == EINTR)
955 continue;
Willy Tarreau843b7cb2018-07-13 10:54:26 +0200956 if (len2 != trash.data)
Bertrand Jacquin93b227d2016-06-04 15:11:10 +0100957 goto recv_abort;
958 } while (0);
959
960 conn->flags &= ~flag;
961 return 1;
962
963 missing:
964 /* Missing data. Since we're using MSG_PEEK, we can only poll again if
965 * we have not read anything. Otherwise we need to fail because we won't
966 * be able to poll anymore.
967 */
968 conn->err_code = CO_ER_CIP_TRUNCATED;
969 goto fail;
970
971 bad_magic:
972 conn->err_code = CO_ER_CIP_BAD_MAGIC;
973 goto fail;
974
975 recv_abort:
976 conn->err_code = CO_ER_CIP_ABORT;
977 conn->flags |= CO_FL_SOCK_RD_SH | CO_FL_SOCK_WR_SH;
978 goto fail;
979
980 fail:
981 __conn_sock_stop_both(conn);
982 conn->flags |= CO_FL_ERROR;
983 return 0;
984}
985
Ilya Shipitsinca56fce2018-09-15 00:50:05 +0500986/* Note: <remote> is explicitly allowed to be NULL */
David Safb76832014-05-08 23:42:08 -0400987int make_proxy_line(char *buf, int buf_len, struct server *srv, struct connection *remote)
988{
989 int ret = 0;
990
991 if (srv && (srv->pp_opts & SRV_PP_V2)) {
992 ret = make_proxy_line_v2(buf, buf_len, srv, remote);
993 }
994 else {
995 if (remote)
996 ret = make_proxy_line_v1(buf, buf_len, &remote->addr.from, &remote->addr.to);
997 else
998 ret = make_proxy_line_v1(buf, buf_len, NULL, NULL);
999 }
1000
1001 return ret;
1002}
1003
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001004/* Makes a PROXY protocol line from the two addresses. The output is sent to
1005 * buffer <buf> for a maximum size of <buf_len> (including the trailing zero).
1006 * It returns the number of bytes composing this line (including the trailing
1007 * LF), or zero in case of failure (eg: not enough space). It supports TCP4,
Willy Tarreau2e1401a2013-10-01 11:41:55 +02001008 * TCP6 and "UNKNOWN" formats. If any of <src> or <dst> is null, UNKNOWN is
1009 * emitted as well.
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001010 */
David Safb76832014-05-08 23:42:08 -04001011int make_proxy_line_v1(char *buf, int buf_len, struct sockaddr_storage *src, struct sockaddr_storage *dst)
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001012{
1013 int ret = 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001014 char * protocol;
1015 char src_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
1016 char dst_str[MAX(INET_ADDRSTRLEN, INET6_ADDRSTRLEN)];
1017 in_port_t src_port;
1018 in_port_t dst_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001019
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001020 if ( !src
1021 || !dst
1022 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
1023 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
1024 /* unknown family combination */
1025 ret = snprintf(buf, buf_len, "PROXY UNKNOWN\r\n");
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001026 if (ret >= buf_len)
1027 return 0;
1028
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001029 return ret;
1030 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001031
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001032 /* IPv4 for both src and dst */
1033 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1034 protocol = "TCP4";
1035 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)src)->sin_addr, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001036 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001037 src_port = ((struct sockaddr_in *)src)->sin_port;
1038 if (!inet_ntop(AF_INET, &((struct sockaddr_in *)dst)->sin_addr, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001039 return 0;
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001040 dst_port = ((struct sockaddr_in *)dst)->sin_port;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001041 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001042 /* IPv6 for at least one of src and dst */
1043 else {
1044 struct in6_addr tmp;
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001045
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001046 protocol = "TCP6";
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001047
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001048 if (src->ss_family == AF_INET) {
1049 /* Convert src to IPv6 */
1050 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1051 src_port = ((struct sockaddr_in *)src)->sin_port;
1052 }
1053 else {
1054 tmp = ((struct sockaddr_in6 *)src)->sin6_addr;
1055 src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1056 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001057
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001058 if (!inet_ntop(AF_INET6, &tmp, src_str, sizeof(src_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001059 return 0;
1060
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001061 if (dst->ss_family == AF_INET) {
1062 /* Convert dst to IPv6 */
1063 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1064 dst_port = ((struct sockaddr_in *)dst)->sin_port;
1065 }
1066 else {
1067 tmp = ((struct sockaddr_in6 *)dst)->sin6_addr;
1068 dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1069 }
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001070
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001071 if (!inet_ntop(AF_INET6, &tmp, dst_str, sizeof(dst_str)))
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001072 return 0;
1073 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001074
1075 ret = snprintf(buf, buf_len, "PROXY %s %s %s %u %u\r\n", protocol, src_str, dst_str, ntohs(src_port), ntohs(dst_port));
1076 if (ret >= buf_len)
1077 return 0;
1078
Willy Tarreaue1e4a612012-10-05 00:10:55 +02001079 return ret;
1080}
David Safb76832014-05-08 23:42:08 -04001081
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001082static int make_tlv(char *dest, int dest_len, char type, uint16_t length, const char *value)
David Safb76832014-05-08 23:42:08 -04001083{
1084 struct tlv *tlv;
1085
1086 if (!dest || (length + sizeof(*tlv) > dest_len))
1087 return 0;
1088
1089 tlv = (struct tlv *)dest;
1090
1091 tlv->type = type;
1092 tlv->length_hi = length >> 8;
1093 tlv->length_lo = length & 0x00ff;
1094 memcpy(tlv->value, value, length);
1095 return length + sizeof(*tlv);
1096}
David Safb76832014-05-08 23:42:08 -04001097
Ilya Shipitsinca56fce2018-09-15 00:50:05 +05001098/* Note: <remote> is explicitly allowed to be NULL */
David Safb76832014-05-08 23:42:08 -04001099int make_proxy_line_v2(char *buf, int buf_len, struct server *srv, struct connection *remote)
1100{
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001101 const char pp2_signature[] = PP2_SIGNATURE;
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001102 void *tlv_crc32c_p = NULL;
David Safb76832014-05-08 23:42:08 -04001103 int ret = 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001104 struct proxy_hdr_v2 *hdr = (struct proxy_hdr_v2 *)buf;
Vincent Bernat6e615892016-05-18 16:17:44 +02001105 struct sockaddr_storage null_addr = { .ss_family = 0 };
David Safb76832014-05-08 23:42:08 -04001106 struct sockaddr_storage *src = &null_addr;
1107 struct sockaddr_storage *dst = &null_addr;
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001108 const char *value;
1109 int value_len;
David Safb76832014-05-08 23:42:08 -04001110
1111 if (buf_len < PP2_HEADER_LEN)
1112 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001113 memcpy(hdr->sig, pp2_signature, PP2_SIGNATURE_LEN);
David Safb76832014-05-08 23:42:08 -04001114
1115 if (remote) {
1116 src = &remote->addr.from;
1117 dst = &remote->addr.to;
1118 }
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001119
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001120 /* At least one of src or dst is not of AF_INET or AF_INET6 */
1121 if ( !src
1122 || !dst
1123 || (src->ss_family != AF_INET && src->ss_family != AF_INET6)
1124 || (dst->ss_family != AF_INET && dst->ss_family != AF_INET6)) {
David Safb76832014-05-08 23:42:08 -04001125 if (buf_len < PP2_HDR_LEN_UNSPEC)
1126 return 0;
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001127 hdr->ver_cmd = PP2_VERSION | PP2_CMD_LOCAL;
1128 hdr->fam = PP2_FAM_UNSPEC | PP2_TRANS_UNSPEC;
David Safb76832014-05-08 23:42:08 -04001129 ret = PP2_HDR_LEN_UNSPEC;
1130 }
Tim Duesterhus7fec0212018-07-27 18:46:13 +02001131 else {
1132 /* IPv4 for both src and dst */
1133 if (src->ss_family == AF_INET && dst->ss_family == AF_INET) {
1134 if (buf_len < PP2_HDR_LEN_INET)
1135 return 0;
1136 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1137 hdr->fam = PP2_FAM_INET | PP2_TRANS_STREAM;
1138 hdr->addr.ip4.src_addr = ((struct sockaddr_in *)src)->sin_addr.s_addr;
1139 hdr->addr.ip4.src_port = ((struct sockaddr_in *)src)->sin_port;
1140 hdr->addr.ip4.dst_addr = ((struct sockaddr_in *)dst)->sin_addr.s_addr;
1141 hdr->addr.ip4.dst_port = ((struct sockaddr_in *)dst)->sin_port;
1142 ret = PP2_HDR_LEN_INET;
1143 }
1144 /* IPv6 for at least one of src and dst */
1145 else {
1146 struct in6_addr tmp;
1147
1148 if (buf_len < PP2_HDR_LEN_INET6)
1149 return 0;
1150 hdr->ver_cmd = PP2_VERSION | PP2_CMD_PROXY;
1151 hdr->fam = PP2_FAM_INET6 | PP2_TRANS_STREAM;
1152 if (src->ss_family == AF_INET) {
1153 v4tov6(&tmp, &((struct sockaddr_in *)src)->sin_addr);
1154 memcpy(hdr->addr.ip6.src_addr, &tmp, 16);
1155 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1156 }
1157 else {
1158 memcpy(hdr->addr.ip6.src_addr, &((struct sockaddr_in6 *)src)->sin6_addr, 16);
1159 hdr->addr.ip6.src_port = ((struct sockaddr_in6 *)src)->sin6_port;
1160 }
1161 if (dst->ss_family == AF_INET) {
1162 v4tov6(&tmp, &((struct sockaddr_in *)dst)->sin_addr);
1163 memcpy(hdr->addr.ip6.dst_addr, &tmp, 16);
1164 hdr->addr.ip6.src_port = ((struct sockaddr_in *)src)->sin_port;
1165 }
1166 else {
1167 memcpy(hdr->addr.ip6.dst_addr, &((struct sockaddr_in6 *)dst)->sin6_addr, 16);
1168 hdr->addr.ip6.dst_port = ((struct sockaddr_in6 *)dst)->sin6_port;
1169 }
1170
1171 ret = PP2_HDR_LEN_INET6;
1172 }
1173 }
David Safb76832014-05-08 23:42:08 -04001174
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001175 if (srv->pp_opts & SRV_PP_V2_CRC32C) {
1176 uint32_t zero_crc32c = 0;
1177 if ((buf_len - ret) < sizeof(struct tlv))
1178 return 0;
1179 tlv_crc32c_p = (void *)((struct tlv *)&buf[ret])->value;
1180 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_CRC32C, sizeof(zero_crc32c), (const char *)&zero_crc32c);
1181 }
1182
Ilya Shipitsinca56fce2018-09-15 00:50:05 +05001183 if (remote && conn_get_alpn(remote, &value, &value_len)) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001184 if ((buf_len - ret) < sizeof(struct tlv))
1185 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001186 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_ALPN, value_len, value);
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001187 }
1188
David Safb76832014-05-08 23:42:08 -04001189#ifdef USE_OPENSSL
Emmanuel Hocdet253c3b72018-02-01 18:29:59 +01001190 if (srv->pp_opts & SRV_PP_V2_AUTHORITY) {
1191 value = ssl_sock_get_sni(remote);
1192 if (value) {
1193 if ((buf_len - ret) < sizeof(struct tlv))
1194 return 0;
1195 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_AUTHORITY, strlen(value), value);
1196 }
1197 }
1198
David Safb76832014-05-08 23:42:08 -04001199 if (srv->pp_opts & SRV_PP_V2_SSL) {
Emmanuel Hocdet404d9782017-10-24 10:55:14 +02001200 struct tlv_ssl *tlv;
1201 int ssl_tlv_len = 0;
David Safb76832014-05-08 23:42:08 -04001202 if ((buf_len - ret) < sizeof(struct tlv_ssl))
1203 return 0;
1204 tlv = (struct tlv_ssl *)&buf[ret];
1205 memset(tlv, 0, sizeof(struct tlv_ssl));
1206 ssl_tlv_len += sizeof(struct tlv_ssl);
1207 tlv->tlv.type = PP2_TYPE_SSL;
1208 if (ssl_sock_is_ssl(remote)) {
1209 tlv->client |= PP2_CLIENT_SSL;
Emmanuel Hocdet01da5712017-10-13 16:59:49 +02001210 value = ssl_sock_get_proto_version(remote);
David Safb76832014-05-08 23:42:08 -04001211 if (value) {
Emmanuel Hocdet8c0c34b2018-02-28 12:02:14 +01001212 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len-ret-ssl_tlv_len), PP2_SUBTYPE_SSL_VERSION, strlen(value), value);
David Safb76832014-05-08 23:42:08 -04001213 }
Dave McCowan328fb582014-07-30 10:39:13 -04001214 if (ssl_sock_get_cert_used_sess(remote)) {
1215 tlv->client |= PP2_CLIENT_CERT_SESS;
David Safb76832014-05-08 23:42:08 -04001216 tlv->verify = htonl(ssl_sock_get_verify_result(remote));
Dave McCowan328fb582014-07-30 10:39:13 -04001217 if (ssl_sock_get_cert_used_conn(remote))
1218 tlv->client |= PP2_CLIENT_CERT_CONN;
David Safb76832014-05-08 23:42:08 -04001219 }
1220 if (srv->pp_opts & SRV_PP_V2_SSL_CN) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001221 struct buffer *cn_trash = get_trash_chunk();
Willy Tarreau3b9a0c92014-07-19 06:37:33 +02001222 if (ssl_sock_get_remote_common_name(remote, cn_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001223 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CN,
1224 cn_trash->data,
1225 cn_trash->area);
David Safb76832014-05-08 23:42:08 -04001226 }
1227 }
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001228 if (srv->pp_opts & SRV_PP_V2_SSL_KEY_ALG) {
Willy Tarreau83061a82018-07-13 11:56:34 +02001229 struct buffer *pkey_trash = get_trash_chunk();
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001230 if (ssl_sock_get_pkey_algo(remote, pkey_trash) > 0) {
Willy Tarreau843b7cb2018-07-13 10:54:26 +02001231 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_KEY_ALG,
1232 pkey_trash->data,
1233 pkey_trash->area);
Emmanuel Hocdetfa8d0f12018-02-01 15:53:52 +01001234 }
1235 }
1236 if (srv->pp_opts & SRV_PP_V2_SSL_SIG_ALG) {
1237 value = ssl_sock_get_cert_sig(remote);
1238 if (value) {
1239 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_SIG_ALG, strlen(value), value);
1240 }
1241 }
1242 if (srv->pp_opts & SRV_PP_V2_SSL_CIPHER) {
1243 value = ssl_sock_get_cipher_name(remote);
1244 if (value) {
1245 ssl_tlv_len += make_tlv(&buf[ret+ssl_tlv_len], (buf_len - ret - ssl_tlv_len), PP2_SUBTYPE_SSL_CIPHER, strlen(value), value);
1246 }
1247 }
David Safb76832014-05-08 23:42:08 -04001248 }
1249 tlv->tlv.length_hi = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) >> 8;
1250 tlv->tlv.length_lo = (uint16_t)(ssl_tlv_len - sizeof(struct tlv)) & 0x00ff;
1251 ret += ssl_tlv_len;
1252 }
1253#endif
1254
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001255#ifdef CONFIG_HAP_NS
1256 if (remote && (remote->proxy_netns)) {
1257 if ((buf_len - ret) < sizeof(struct tlv))
1258 return 0;
Emmanuel Hocdet571c7ac2017-10-31 18:24:05 +01001259 ret += make_tlv(&buf[ret], (buf_len - ret), PP2_TYPE_NETNS, remote->proxy_netns->name_len, remote->proxy_netns->node.key);
KOVACS Krisztianb3e54fe2014-11-17 15:11:45 +01001260 }
1261#endif
1262
Willy Tarreau8fccfa22014-06-14 08:28:06 +02001263 hdr->len = htons((uint16_t)(ret - PP2_HEADER_LEN));
David Safb76832014-05-08 23:42:08 -04001264
Emmanuel Hocdet4399c752018-02-05 15:26:43 +01001265 if (tlv_crc32c_p) {
1266 write_u32(tlv_crc32c_p, htonl(hash_crc32c(buf, ret)));
1267 }
1268
David Safb76832014-05-08 23:42:08 -04001269 return ret;
1270}
Emeric Brun4f603012017-01-05 15:11:44 +01001271
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001272/* return the major HTTP version as 1 or 2 depending on how the request arrived
1273 * before being processed.
1274 */
1275static int
1276smp_fetch_fc_http_major(const struct arg *args, struct sample *smp, const char *kw, void *private)
1277{
1278 struct connection *conn = objt_conn(smp->sess->origin);
1279
1280 smp->data.type = SMP_T_SINT;
1281 smp->data.u.sint = (conn && strcmp(conn_get_mux_name(conn), "H2") == 0) ? 2 : 1;
1282 return 1;
1283}
1284
Emeric Brun4f603012017-01-05 15:11:44 +01001285/* fetch if the received connection used a PROXY protocol header */
1286int smp_fetch_fc_rcvd_proxy(const struct arg *args, struct sample *smp, const char *kw, void *private)
1287{
1288 struct connection *conn;
1289
1290 conn = objt_conn(smp->sess->origin);
1291 if (!conn)
1292 return 0;
1293
1294 if (!(conn->flags & CO_FL_CONNECTED)) {
1295 smp->flags |= SMP_F_MAY_CHANGE;
1296 return 0;
1297 }
1298
1299 smp->flags = 0;
1300 smp->data.type = SMP_T_BOOL;
1301 smp->data.u.sint = (conn->flags & CO_FL_RCVD_PROXY) ? 1 : 0;
1302
1303 return 1;
1304}
1305
1306/* Note: must not be declared <const> as its list will be overwritten.
1307 * Note: fetches that may return multiple types must be declared as the lowest
1308 * common denominator, the type that can be casted into all other ones. For
1309 * instance v4/v6 must be declared v4.
1310 */
1311static struct sample_fetch_kw_list sample_fetch_keywords = {ILH, {
Willy Tarreau60ca10a2017-08-18 15:26:54 +02001312 { "fc_http_major", smp_fetch_fc_http_major, 0, NULL, SMP_T_SINT, SMP_USE_L4CLI },
Emeric Brun4f603012017-01-05 15:11:44 +01001313 { "fc_rcvd_proxy", smp_fetch_fc_rcvd_proxy, 0, NULL, SMP_T_BOOL, SMP_USE_L4CLI },
1314 { /* END */ },
1315}};
1316
1317
1318__attribute__((constructor))
1319static void __connection_init(void)
1320{
1321 sample_register_fetches(&sample_fetch_keywords);
1322}